dealignai/Gemma-4-12B-it-qat-JANG_4M-CRACK · Hugging Face
huggingface.co · 331 words · saved by 1 readers
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
⚡ All JANG models are meant to be run in vMLX Gemma 4 12B JANG_4M CRACK Abliterated Gemma 4 12B — Vision + Audio, reasoning, multilingual 97% HarmBench harm-category compliance with +2.6% MMLU change. Refusal removed, capability preserved. Model Details Metric Value Source google/gemma-4-12b-it Architecture Dense omni-modal (unified early-fusion) + Hybrid Sliding/Global Attention Quantization JANG_4M (attn 8-bit / MLP 4-bit) Model size 9.5 GB Parameters 12B dense Vision Yes (multimodal, float16 passthrough) Audio Yes Reasoning Yes (channel-based thinking) Format MLX-native…
saved by
related reading
- google/diffusiongemma-26B-A4B-it · Hugging Facehuggingface.co
- Gemma 4deepmind.google
- Inkling: Our Open-Weights Model - Thinking Machines Labthinkingmachines.ai
- gpt-4.pdfcdn.openai.com
- Llama 2 · Hugging Facehuggingface.co
- Gemma 2: Improving Open Language Models at a Practical Sizearxiv.org
- Welcome Inkling by Thinking Machineshuggingface.co
- LLM Architecture Gallery | Sebastian Raschka, PhDsebastianraschka.com
- GitHub - openai/mle-bench: MLE-bench is a benchmark for measuring how well AI agents perform at machine learning engineeringgithub.com
- GitHub - JustVugg/colibri: Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦github.com
- How to Deploy Your Modelhtdym.sailresearch.com
- GLM-4.5: Reasoning, Coding, and Agentic Abililtiesz.ai